diff --git a/src/ptbench/engine/visualizer.py b/src/ptbench/engine/visualizer.py
index 87ceca5998b5bba39a8e38c7cd2cdf42e2c7763f..46f22e02e0842c3073b99a7fba862ac0abd10c3f 100644
--- a/src/ptbench/engine/visualizer.py
+++ b/src/ptbench/engine/visualizer.py
@@ -9,11 +9,9 @@ import cv2
import numpy as np
import pandas as pd
-from PIL import Image
+from torchvision.transforms.functional import to_pil_image
from tqdm import tqdm
-from torchvision.transforms.functional import to_pil_image, to_tensor
-
from ..utils.cam_utils import (
draw_boxes_on_image,
draw_largest_component_bbox_on_image,
diff --git a/src/ptbench/scripts/compare_vis.py b/src/ptbench/scripts/compare_vis.py
index 6cee4b9366debc7ee46348b615bbcd484c76315e..bade26b87c4e53b2d18bfe1c9450b8937b8b6b08 100644
--- a/src/ptbench/scripts/compare_vis.py
+++ b/src/ptbench/scripts/compare_vis.py
@@ -26,6 +26,7 @@ def _sorting_rule(folder_name):
else: # Everything else will be sorted alphabetically after fullgrad and before randomcam
return (3, folder_name)
+
def _get_images_from_directory(dir_path):
image_files = []
for root, _, files in os.walk(dir_path):
@@ -34,6 +35,7 @@ def _get_images_from_directory(dir_path):
image_files.append(os.path.join(root, file))
return image_files
+
@click.command(
epilog="""Examples:
@@ -133,13 +135,19 @@ def compare_vis(input_folder, output_folder) -> None:
os.makedirs(output_directory, exist_ok=True)
# Use a set (unordered collection of unique elements) for efficient membership tests
- image_names = set([os.path.basename(img) for img in _get_images_from_directory(comparison_folders[0])])
+ image_names = {
+ os.path.basename(img)
+ for img in _get_images_from_directory(comparison_folders[0])
+ }
# Only keep image names that exist in all folders
for folder in comparison_folders[1:]:
# This is basically an intersection-check of contents of different folders
# Images that don't exist in all folders are removed from the set
- image_names &= set([os.path.basename(img) for img in _get_images_from_directory(folder)])
+ image_names &= {
+ os.path.basename(img)
+ for img in _get_images_from_directory(folder)
+ }
if not image_names:
raise ValueError("No common images found in the folders.")
@@ -165,7 +173,11 @@ def compare_vis(input_folder, output_folder) -> None:
)
axs = axs.ravel()
for i, folder in enumerate(comparison_folders):
- image_path = [img for img in _get_images_from_directory(folder) if os.path.basename(img) == image_name][0]
+ image_path = [
+ img
+ for img in _get_images_from_directory(folder)
+ if os.path.basename(img) == image_name
+ ][0]
try:
img = cv2.imread(image_path)
img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
diff --git a/tests/test_cam_utils.py b/tests/test_cam_utils.py
index 1435d8ac9d3ec294e969dbae737884c079689027..2d3273d55c08aa9e5d7ba5da6293abae959553d6 100644
--- a/tests/test_cam_utils.py
+++ b/tests/test_cam_utils.py
@@ -217,8 +217,20 @@ def test_draw_boxes_on_image():
img_with_boxes2 = np.zeros((256, 256, 3), dtype=np.uint8)
img_with_boxes3 = np.zeros((512, 512, 3), dtype=np.uint8)
bboxes = [
- [torch.tensor(1),torch.tensor(10),torch.tensor(10),torch.tensor(20),torch.tensor(20)],
- [torch.tensor(1),torch.tensor(30),torch.tensor(30),torch.tensor(40),torch.tensor(40)]
+ [
+ torch.tensor(1),
+ torch.tensor(10),
+ torch.tensor(10),
+ torch.tensor(20),
+ torch.tensor(20),
+ ],
+ [
+ torch.tensor(1),
+ torch.tensor(30),
+ torch.tensor(30),
+ torch.tensor(40),
+ torch.tensor(40),
+ ],
]
result = draw_boxes_on_image(img_with_boxes, bboxes)
diff --git a/tests/test_saliencymap_evaluator.py b/tests/test_saliencymap_evaluator.py
index 1b32f283883130acee3c911e101656fdee800574..848812402405eaa9bf6e6bc86460c80b4ca12ba2 100644
--- a/tests/test_saliencymap_evaluator.py
+++ b/tests/test_saliencymap_evaluator.py
@@ -15,8 +15,20 @@ from ptbench.engine.saliencymap_evaluator import (
def test_compute_max_iou_and_ioda():
detected_box = (10, 10, 100, 100)
- gt_box_dict = [torch.tensor(1),torch.tensor(50),torch.tensor(50),torch.tensor(50),torch.tensor(50)]
- gt_box_dict2 = [torch.tensor(1),torch.tensor(20),torch.tensor(20),torch.tensor(60),torch.tensor(60)]
+ gt_box_dict = [
+ torch.tensor(1),
+ torch.tensor(50),
+ torch.tensor(50),
+ torch.tensor(50),
+ torch.tensor(50),
+ ]
+ gt_box_dict2 = [
+ torch.tensor(1),
+ torch.tensor(20),
+ torch.tensor(20),
+ torch.tensor(60),
+ torch.tensor(60),
+ ]
gt_boxes = [gt_box_dict, gt_box_dict2]
iou, ioda = _compute_max_iou_and_ioda(detected_box, gt_boxes)
@@ -30,7 +42,13 @@ def test_compute_max_iou_and_ioda():
def test_compute_max_iou_and_ioda_zero_detected_area():
detected_box = (10, 10, 0, 0)
- gt_box_dict = [torch.tensor(1),torch.tensor(50),torch.tensor(50),torch.tensor(50),torch.tensor(50)]
+ gt_box_dict = [
+ torch.tensor(1),
+ torch.tensor(50),
+ torch.tensor(50),
+ torch.tensor(50),
+ torch.tensor(50),
+ ]
gt_boxes = [gt_box_dict]
iou, ioda = _compute_max_iou_and_ioda(detected_box, gt_boxes)
@@ -42,7 +60,13 @@ def test_compute_max_iou_and_ioda_zero_detected_area():
def test_compute_max_iou_and_ioda_zero_gt_area():
detected_box = (10, 10, 100, 100)
- gt_box_dict = [torch.tensor(1),torch.tensor(50),torch.tensor(50),torch.tensor(0),torch.tensor(0)]
+ gt_box_dict = [
+ torch.tensor(1),
+ torch.tensor(50),
+ torch.tensor(50),
+ torch.tensor(0),
+ torch.tensor(0),
+ ]
gt_boxes = [gt_box_dict]
iou, ioda = _compute_max_iou_and_ioda(detected_box, gt_boxes)
@@ -54,7 +78,13 @@ def test_compute_max_iou_and_ioda_zero_gt_area():
def test_compute_max_iou_and_ioda_zero_intersection():
detected_box = (10, 10, 100, 100)
- gt_box_dict = [torch.tensor(1),torch.tensor(0),torch.tensor(0),torch.tensor(5),torch.tensor(5)]
+ gt_box_dict = [
+ torch.tensor(1),
+ torch.tensor(0),
+ torch.tensor(0),
+ torch.tensor(5),
+ torch.tensor(5),
+ ]
gt_boxes = [gt_box_dict]
iou, ioda = _compute_max_iou_and_ioda(detected_box, gt_boxes)
@@ -65,8 +95,20 @@ def test_compute_max_iou_and_ioda_zero_intersection():
def test_compute_simultaneous_iou_and_ioda():
detected_box = (10, 10, 100, 100)
- gt_box_dict1 = [torch.tensor(1),torch.tensor(50),torch.tensor(50),torch.tensor(50),torch.tensor(50)]
- gt_box_dict2 = [torch.tensor(1),torch.tensor(70),torch.tensor(70),torch.tensor(30),torch.tensor(30)]
+ gt_box_dict1 = [
+ torch.tensor(1),
+ torch.tensor(50),
+ torch.tensor(50),
+ torch.tensor(50),
+ torch.tensor(50),
+ ]
+ gt_box_dict2 = [
+ torch.tensor(1),
+ torch.tensor(70),
+ torch.tensor(70),
+ torch.tensor(30),
+ torch.tensor(30),
+ ]
gt_boxes = [gt_box_dict1, gt_box_dict2]
iou, ioda = _compute_simultaneous_iou_and_ioda(detected_box, gt_boxes)
@@ -80,7 +122,13 @@ def test_compute_avg_saliency_focus():
grayscale_cams2 = np.full((512, 512), 0.5)
grayscale_cams3 = np.zeros((256, 256))
grayscale_cams3[50:75, 50:100] = 1
- gt_box_dict = [torch.tensor(1),torch.tensor(50),torch.tensor(50),torch.tensor(50),torch.tensor(50)]
+ gt_box_dict = [
+ torch.tensor(1),
+ torch.tensor(50),
+ torch.tensor(50),
+ torch.tensor(50),
+ torch.tensor(50),
+ ]
gt_boxes = [gt_box_dict]
avg_saliency_focus = _compute_avg_saliency_focus(gt_boxes, grayscale_cams)
@@ -94,7 +142,13 @@ def test_compute_avg_saliency_focus():
def test_compute_avg_saliency_focus_no_activations():
grayscale_cams = np.zeros((200, 200))
- gt_box_dict = [torch.tensor(1),torch.tensor(50),torch.tensor(50),torch.tensor(50),torch.tensor(50)]
+ gt_box_dict = [
+ torch.tensor(1),
+ torch.tensor(50),
+ torch.tensor(50),
+ torch.tensor(50),
+ torch.tensor(50),
+ ]
gt_boxes = [gt_box_dict]
avg_saliency_focus = _compute_avg_saliency_focus(gt_boxes, grayscale_cams)
@@ -104,7 +158,13 @@ def test_compute_avg_saliency_focus_no_activations():
def test_compute_avg_saliency_focus_zero_gt_area():
grayscale_cams = np.ones((200, 200))
- gt_box_dict = [torch.tensor(1),torch.tensor(50),torch.tensor(50),torch.tensor(0),torch.tensor(0)]
+ gt_box_dict = [
+ torch.tensor(1),
+ torch.tensor(50),
+ torch.tensor(50),
+ torch.tensor(0),
+ torch.tensor(0),
+ ]
gt_boxes = [gt_box_dict]
avg_saliency_focus = _compute_avg_saliency_focus(gt_boxes, grayscale_cams)
@@ -117,7 +177,13 @@ def test_compute_proportional_energy():
grayscale_cams2 = np.full((512, 512), 0.5)
grayscale_cams3 = np.zeros((512, 512))
grayscale_cams3[100:200, 100:200] = 1
- gt_box_dict = [torch.tensor(1),torch.tensor(50),torch.tensor(50),torch.tensor(100),torch.tensor(100)]
+ gt_box_dict = [
+ torch.tensor(1),
+ torch.tensor(50),
+ torch.tensor(50),
+ torch.tensor(100),
+ torch.tensor(100),
+ ]
gt_boxes = [gt_box_dict]
proportional_energy = _compute_proportional_energy(gt_boxes, grayscale_cams)
@@ -135,7 +201,13 @@ def test_compute_proportional_energy():
def test_compute_proportional_energy_no_activations():
grayscale_cams = np.zeros((200, 200))
- gt_box_dict = [torch.tensor(1),torch.tensor(50),torch.tensor(50),torch.tensor(50),torch.tensor(50)]
+ gt_box_dict = [
+ torch.tensor(1),
+ torch.tensor(50),
+ torch.tensor(50),
+ torch.tensor(50),
+ torch.tensor(50),
+ ]
gt_boxes = [gt_box_dict]
proportional_energy = _compute_proportional_energy(gt_boxes, grayscale_cams)
@@ -145,7 +217,13 @@ def test_compute_proportional_energy_no_activations():
def test_compute_proportional_energy_no_gt_box():
grayscale_cams = np.ones((200, 200))
- gt_box_dict = [torch.tensor(1),torch.tensor(0),torch.tensor(0),torch.tensor(0),torch.tensor(0)]
+ gt_box_dict = [
+ torch.tensor(1),
+ torch.tensor(0),
+ torch.tensor(0),
+ torch.tensor(0),
+ torch.tensor(0),
+ ]
gt_boxes = [gt_box_dict]
proportional_energy = _compute_proportional_energy(gt_boxes, grayscale_cams)
@@ -156,7 +234,13 @@ def test_compute_proportional_energy_no_gt_box():
def test_calculate_localization_metrics():
grayscale_cams = np.zeros((200, 200))
detected_box = (10, 10, 100, 100)
- gt_box_dict = [torch.tensor(1),torch.tensor(50),torch.tensor(50),torch.tensor(50),torch.tensor(50)]
+ gt_box_dict = [
+ torch.tensor(1),
+ torch.tensor(50),
+ torch.tensor(50),
+ torch.tensor(50),
+ torch.tensor(50),
+ ]
ground_truth_box = [gt_box_dict]
(